Author(s)

SATOSHI ANZAI, SO KAZAMA

Abstract

During floods, some people approach flooded riversides out of curiosity, elevating their risk of being swept away by floodwaters. Social media was used to investigate actual patterns of this high-risk behaviour and to understand its temporal and spatial distribution. Among various social media, Instagram was chosen for its real-time nature and low-noise characteristics. The selected study areas included Tama river, Kinu river, Edo river, Tone river, Sagami river, and Hirose river in Japan, areas that were all affected by heavy rain from the 7th to the 12th of September 2015. Data was collected by Instagram API detecting hash-tag search of the river names, such as “#Tama river”. The collected data was assessed by time series analysis, text analysis, and image analysis. The results indicated that usage of the relevant hashtags increased following a flood warning being issued. Additionally, the number of posts following the flood warning was influenced by population characteristics along rivers and warning issue time. Thirdly, results revealed that many people continued to remain on the riverbanks despite their awareness of the increasing risk.

Keywords

big data, Instagram, text analysis